Longitudinal Analysis (Model-based approach)

Generalized Least Squares

---

Description

Tech

• Regression modeling strategy
  • we use a property to characterise a response within groups
  • after specifying a regression structure, we optimize the model by some criteria
    • AIC, BIC … for time-consuming models + large-scaled data
    • AUROC, AUPRC … for prediction-focused + smaller-size data
  • if assumptions are satisfied with the optimized model, we are allowed to do inference on estimated responce and effects.
• Longitudinal Data (serial data/repeated measures) Analysis
  • Model-based
    • General Least Squares (GLS)
    • Mixed Effect Model
    • Bayesian hierarchical models
  • Data-based
    • Summary Measures

Medical

• 4 primary groups
  • non_pos: postive episodes from no-transplant patients
  • non_neg: test-negative and baseline time windows from ‘pure negative’ no-transplant patients
  • txp_pos: postive episodes from transplant patients
  • txp_neg: test-negative and baseline time windows from ‘pure negative’ transplant patients
• An episode
  • from 24 hours before to 24 hours after a blood stream acquisition
• Questions
  1. What’s the response trend of a feature during one ‘episode’ within groups?
  2. What’s the difference of response trend of a feature during one ‘episode’ across groups?

---

Report

• Assumption-satisfied Variables
  • Temp
  • Resp
  • Pulse
  • SBP
  • DBP
  • POTASSIUM
  • ALBUMIN
  • CALCIUM
  • SODIUM
  • WHITE_BLOOD_CELL_COUNT
  • PHOSPHORUS
  • PLATELET_COUNT
  • CO2
  • PCO2
  • CHLORIDE
  • BLOOD_UREA_NITROGEN
  • MAGNESIUM
• model assumption checking plots
• partial effect plots
• contrast (subtract) inference plots
  • by subtracting estimated response from 2 groups and comparing the confidence interval with 0, we can do inference (example - Temperature)
    • In no-transplant group, our data suggest a significant increase in temperature in postive episodes than control (negative+baseline) episodes
    • In transplant group, our data don’t suggest a significant increase in positive episodes than control (negative-baseline) episodes

---

---

Temp mean response

• Check model assumptions
• Partiel effect and contrast inference ***

Resp mean response

• Check model assumptions
• Partiel effect and contrast inference ***

Pulse mean response

• Check model assumptions
• Partiel effect and contrast inference ***

SBP mean response

• Check model assumptions
• Partiel effect and contrast inference ***

DBP mean response

• Check model assumptions
• Partiel effect and contrast inference ***

POTASSIUM mean response

• Check model assumptions
• Partiel effect and contrast inference ***

ALBUMIN mean response

• Check model assumptions
• Partiel effect and contrast inference ***

CALCIUM mean response

• Check model assumptions
• Partiel effect and contrast inference ***

SODIUM mean response

• Check model assumptions
• Partiel effect and contrast inference ***

WHITE_BLOOD_CELL_COUNT mean response

• Check model assumptions
• Partiel effect and contrast inference ***

PHOSPHORUS mean response

• Check model assumptions
• Partiel effect and contrast inference ***

PLATELET_COUNT mean response

• Check model assumptions
• Partiel effect and contrast inference ***

CO2 mean response

• Check model assumptions
• Partiel effect and contrast inference ***

PCO2 mean response

• Check model assumptions
• Partiel effect and contrast inference ***

CHLORIDE mean response

• Check model assumptions
• Partiel effect and contrast inference ***

BLOOD_UREA_NITROGEN mean response

• Check model assumptions
• Partiel effect and contrast inference ***

MAGNESIUM mean response

• Check model assumptions
• Partiel effect and contrast inference